Low-resolution palmprint image denoising by generative adversarial networks

Palmprint recognition is a reliable biometric identification method because palmprints contain rich and discriminative features. Low-resolution palmprints have attracted much attention due to their simple acquisition and low computational cost. Many previous works have achieved impressive results. However, we noticed that the performances of these methods declined significantly when there was noise in the palmprint images. Traditional denoising algorithms cannot address multiple types of noise in palmprint images and destroy the orientation information, which is of vital importance for recognition. In this paper, we propose a generative adversarial network (GAN)-based model to cope with this problem. This model is an effective denoising method for low-resolution palmprint images that can handle multiple types of noise and reserve more orientation information. The comparative experimental results demonstrate that our model has reached the status of state-of-the-art image inpainting algorithms with accurate masks. The EER (equal error rate) of the palmprint matching decreased from 10.841% to 1.532% after denoising. Moreover, our method is end-to-end and does not require the additional location information of noise.